Alibaba has just released a real technological bomb: more than 100 new AI models open source code. But for what purpose?
At its annual conference ApsaraAlibaba Cloud revealed more than 100 open source AIunder the name of Qwen 2.5. With this announcement, the Chinese giant also revealed a new infrastructure comprehensive to support the development of artificial intelligence on a large scale.
More than 100 open source AI: why such generosity?
Alibaba Cloudthe cloud computing division of Alibaba Group, has made the source code public of more than 100 AI. These models, known as Qwen 2.5, come in different sizes. They range from 0.5 to 72 billion of parameters.
These AIs are characterized by their increased capacities in mathematics and programming. They support more than 29 languages. They are designed for varied applicationsranging from scientific research to the automotive industry and video games.
Since theirs launch in April 2023Qwen models have already accumulated more than 40 million downloads. And this on specialized platforms such as Hugs face And ModelScope. These models have even given rise to more than 50,000 derived models just on Hugging Face.
According to Jingren ZhouCTO of Alibaba Cloud Intelligence: ” This initiative aims to give more power to developers and companies, regardless of their size. The goal is to improve their ability to leverage AI technologies and promote the growth of the open source community. »
For more efficient AI?
Apart from open source models, Alibaba Cloud also announced update of its proprietary model Qwen-Max. This improved version competes with the best models market leader in language understanding, mathematical reasoning and coding.
As part of this announcement, Alibaba also unveiled a new full-stack infrastructure. The latter is specially designed to meet the increasing demand for computing power for AI.
This infrastructure is integrated innovations for computers, networks and data centers. The goal is to support development and implementation large AI models.
The company also strengthened its multimodal options by introducing a new one text to video conversion template. Integrated into the family Tongyi Wanxiangis this tool capable of producing high quality videos in different visual styles. And this, only fromtext prompt in Chinese and English.
Alibaba Cloud also revealed Qwen2-VLan improved version of his visual language model. This may include videos of more than 20 minutes and manage questions and answers based on these videos.
At the same time, the company launched AI developeran AI assistant based on Qwen. It is intended to help programmers to automate tasks such as requirements analysis, code programming, and error detection and fixing.
Alibaba Cloud is also revolutionizing its infrastructure…
Before support these advances in artificial intelligenceAlibaba Cloud announced several improvements to its infrastructure. Among the latter are:
- CUBE DC 5.0a next-generation data center architecture that improves energy and operational efficiency.
- Alibaba Cloud Open Lakea solution designed to optimize the use of data in generative AI applications.
- PAI AI Schedulera cloud-native planning engine for optimized management of IT resources.
- DMS : OneMeta+OneOps, a unified metadata management platform across multiple cloud environments.
- Enterprise Elastic Compute Service instance (ECS) 9th generation that provides increased performance for various applications.
These Alibaba Cloud updates, including the release of open source models, provide a full support to customers and business partners. They aim for maximize the benefits of newer technologies to create AI applications more efficient, sustainable and inclusive.
- Share the article:
Our blog is powered by readers. When you purchase through links on our site, we may earn an affiliate commission.